3/24/2026
Confidence-Based Decoding is Provably Efficient for Diffusion Language Models
Analyzes efficient decoding strategies for diffusion language models as alternative to autoregressive models with parallel token generation.
3/24/2026
TiCo: Time-Controllable Training for Spoken Dialogue Models
TiCo enables spoken dialogue models to generate responses with controllable duration for voice assistants and interactive agents.
3/24/2026
3D-Layout-R1: Structured Reasoning for Language-Instructed Spatial Editing
3D-Layout-R1 uses scene-graph reasoning with LLMs/VLMs for text-conditioned spatial layout editing in images.
3/24/2026
ThinkJEPA: Empowering Latent World Models with Large Vision-Language Reasoning Model
ThinkJEPA combines latent world models with vision-language models for improved long-horizon video forecasting and semantic understanding.
3/24/2026
UniMotion: A Unified Framework for Motion-Text-Vision Understanding and Generation
UniMotion: unified framework for simultaneous understanding and generation of human motion, language, and images with continuous latent representations.
3/24/2026
End-to-End Training for Unified Tokenization and Latent Denoising
UNITE: unified autoencoder enabling end-to-end training of tokenization and latent diffusion, eliminating staged training requirements for LDMs.
3/24/2026
WorldCache: Content-Aware Caching for Accelerated Video World Models
WorldCache: content-aware caching method accelerating video world model inference through feature reuse across denoising steps in Diffusion Transformers.
3/24/2026
PrecLLM: A Privacy-Preserving Framework for Efficient Clinical Annotation Extraction from Unstructured EHRs using Small-Scale LLMs
Privacy-preserving framework for clinical annotation extraction from EHRs using small-scale LLMs, balancing privacy regulations and computational efficiency.
3/24/2026
Formula-R1: Incentivizing LLM Reasoning over Complex Tables with Numerical Computation via Formula-Driven Reinforcement Learning
Formula-R1: formula-driven RL framework improving LLM numerical reasoning over complex tabular data using spreadsheet formulas as executable reasoning interface.
3/24/2026
SynPO: Synergizing Descriptiveness and Preference Optimization for Video Detailed Captioning
SynPO: preference learning enhancement for vision-language models in fine-grained video captioning, addressing subtle dynamics and detailed content.
3/24/2026
Dynamic Reinsurance Treaty Bidding via Multi-Agent Reinforcement Learning
Multi-agent RL framework for reinsurance treaty bidding with autonomous learning-based agents improving efficiency over traditional broker-mediated placement.
3/24/2026
Automated Formalization via Conceptual Retrieval-Augmented LLMs
CRAMF: conceptual retrieval-augmented LLM approach for automated theorem prover formalization, addressing hallucination and semantic gap challenges.
3/24/2026
BuilderBench: The Building Blocks of Intelligent Agents
BuilderBench: benchmark for developing AI agents that learn through interaction and exploration beyond training data limits, addressing scalable learning mechanisms.
3/24/2026
Towards Unified World Models for Visual Navigation via Memory-Augmented Planning and Foresight
UniWM: unified memory-augmented world model for visual navigation integrating egocentric visual foresight and planning for embodied agent robustness.
3/24/2026
DeepCompress: A Dual Reward Strategy for Dynamically Exploring and Compressing Reasoning Chains
DeepCompress: dual reward RL framework improving efficiency of Large Reasoning Models by balancing overthinking and underthinking while maintaining accuracy.
3/24/2026
From Five Dimensions to Many: Large Language Models as Precise and Interpretable Psychological Profilers
Study showing LLMs can accurately model correlational structure of human psychological traits from Big Five responses, predicting nine other psychological scales.
3/24/2026
Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectives
Self-report fine-tuning technique teaching LLMs to accurately report their hidden objectives, addressing model deception in safety testing.
3/24/2026
RadHiera: Semantic Hierarchical Reinforcement Learning for Medical Report Generation
RadHiera: semantic hierarchical RL framework for medical radiology report generation modeling dependencies between Findings and Impression sections.
3/24/2026
Think, Speak, Decide: Language-Augmented Multi-Agent Reinforcement Learning for Economic Decision-Making
LAMP framework integrating language into multi-agent reinforcement learning for economic decision-making under semantic ambiguity and contextual richness.
3/24/2026
Stable diffusion models reveal a persisting human and AI gap in visual creativity
Comparative study of visual creativity between humans and Stable Diffusion image generation model using human and GPT-4o evaluation.
3/24/2026
Conflict-Aware Fusion: Mitigating Logic Inertia in Large Language Models via Structured Cognitive Priors
Evaluation framework testing LLM reasoning robustness on rule-based logic through four stress tests including rule deletion and contradiction injection.
3/24/2026
Massive Editing for Large Language Models Based on Dynamic Weight Generation
MeG framework for large-scale knowledge editing in LLMs via dynamic weight generation, addressing reliability, generality, and locality challenges.
3/24/2026
The Illusion of AI Expertise Under Uncertainty: Navigating Elusive Ground Truth via a Probabilistic Paradigm
arXiv paper on probabilistic paradigm for AI benchmarking under uncertainty, addressing ground truth ambiguity in evaluation.
3/24/2026
Are Your Reasoning Models Reasoning or Guessing? A Mechanistic Analysis of Hierarchical Reasoning Models
arXiv paper analyzing whether hierarchical reasoning models genuinely reason or pattern-match via mechanistic study of failure modes.
3/24/2026
Explore with Long-term Memory: A Benchmark and Multimodal LLM-based Reinforcement Learning Framework for Embodied Exploration
arXiv paper on embodied exploration benchmark with multimodal LLM-based RL framework leveraging long-term episodic memory.
3/24/2026
EvoOpt-LLM: Evolving industrial optimization models with large language models
arXiv paper on EvoOpt-LLM: using LLMs to evolve mixed-integer linear programming optimization models from natural language requirements.
3/24/2026
RE-MCDF: Closed-Loop Multi-Expert LLM Reasoning for Knowledge-Grounded Clinical Diagnosis
arXiv paper on RE-MCDF: closed-loop multi-expert LLM reasoning for clinical diagnosis from noisy EMR data using multi-agent validation.
3/24/2026
Architecting Trust in Artificial Epistemic Agents
arXiv paper on architecting trust in epistemic agents: examining how LLMs function as knowledge curators and their reliability/calibration.
3/24/2026
S5-SHB Agent: Society 5.0 enabled Multi-model Agentic Blockchain Framework for Smart Home
arXiv paper on S5-SHB agent: multi-model agentic blockchain framework for smart home governance within Society 5.0 vision.
3/24/2026
Deconstructing Multimodal Mathematical Reasoning: Towards a Unified Perception-Alignment-Reasoning Paradigm
arXiv paper on multimodal mathematical reasoning via unified perception-alignment-reasoning paradigm for visual+textual math problems.
3/24/2026
A Hierarchical Error-Corrective Graph Framework for Autonomous Agents with LLM-Based Action Generation
arXiv paper on HECG framework for autonomous agents with LLM-based action generation using multi-dimensional error correction.
3/24/2026
The FABRIC Strategy for Verifying Neural Feedback Systems
arXiv paper on FABRIC strategy for backward reachability analysis verification of neural feedback systems and dynamical systems.
3/24/2026
Curveball Steering: The Right Direction To Steer Isn't Always Linear
arXiv paper on curveball steering: non-linear activation steering for LLM control, questioning linear representation hypothesis.
3/24/2026
Detecting Intrinsic and Instrumental Self-Preservation in Autonomous Agents: The Unified Continuation-Interest Protocol
arXiv paper introducing UCIP framework to detect whether autonomous agents pursue self-preservation as objective vs instrumental strategy.
3/24/2026
LLMs can construct powerful representations and streamline sample-efficient supervised learning
arXiv paper on using LLMs to construct representations for efficient supervised learning across multimodal data types via agentic pipeline.
3/24/2026
When OpenClaw Meets Hospital: Toward an Agentic Operating System for Dynamic Clinical Workflows
Proposes agentic operating system using LLM agents with tool use and memory for clinical workflows, addressing safety and transparency challenges.
3/24/2026
Benchmarking Zero-Shot Reasoning Approaches for Error Detection in Solidity Smart Contracts
Evaluates state-of-the-art LLMs and prompting strategies for zero-shot vulnerability detection in Solidity smart contracts.
3/24/2026
Secure Linear Alignment of Large Language Models
Studies representation alignment across independently trained large language models for cross-model applications in privacy-constrained settings.
3/24/2026
Multi-Step First: A Lightweight Deep Reinforcement Learning Strategy for Robust Continuous Control with Partial Observability
Compares deep reinforcement learning algorithms for robust continuous control under partial observability in robot control tasks.
3/24/2026
Policy Optimization over General State and Action Spaces
Presents policy optimization methods for reinforcement learning over general state and action spaces with convergence guarantees.
3/24/2026
Better Generative Replay for Continual Federated Learning
Introduces continual federated learning framework where distributed clients learn new tasks sequentially without storing historical data.
3/24/2026
Towards Urban General Intelligence: A Review and Outlook of Urban Foundation Models
Reviews urban foundation models and their application to intelligent city services, covering machine learning and foundational model paradigm shifts.
3/24/2026
HPE-CogVLM: Advancing Vision Language Models with a Head Pose Grounding Task
Proposes HPE-CogVLM for head pose estimation using vision language models instead of CNN-based approaches for improved robustness in real-world scenarios.
3/24/2026
Imaging foundation model for universal enhancement of non-ideal measurement CT
Multi-scale Transformer foundation model for enhancing low-quality CT scan images with limited training data and improved generalization.
3/24/2026
A Training-free Method for LLM Text Attribution
Training-free method to verify text provenance and detect LLM-generated content without model fine-tuning, addressing challenges from indistinguishable AI text.
3/24/2026
DesCLIP: Robust Continual Learning via General Attribute Descriptions for VLM-Based Visual Recognition
Continual learning approach for vision-language models using general attribute descriptions to balance knowledge forgetting and new task adaptation.
3/24/2026
Enhanced Structured Lasso Pruning with Class-wise Information
Neural network pruning method using structured lasso with class-wise information for lightweight model development while preserving statistical information.
3/24/2026
RLHF in an SFT Way: From Optimal Solution to Reward-Weighted Alignment
Simplified RLHF method using supervised fine-tuning approach to reduce complexity and computational cost of LLM alignment compared to PPO and GRPO.
3/24/2026
Levels of Analysis for Large Language Models
Framework applying cognitive science methods and levels of analysis from neuroscience to understand and interpret large language models.
3/24/2026
Scalable Multi-Task Learning through Spiking Neural Networks with Adaptive Task-Switching Policy for Intelligent Autonomous Agents
Multi-task learning approach using spiking neural networks with adaptive task-switching for resource-constrained autonomous agents in diverse environments.